Hypernymy is a semantic relation, expressing the “is-a” relation between a concept and its instances. Such relations are building blocks for large-scale taxonomies, ontologies and knowledge graphs. Recently, much progress has been made for hypernymy prediction in English using textual patterns and/or distributional representations. However, applying such techniques to other languages is challenging due to the high language dependency of these methods and the lack of large training datasets of lower-resourced languages. In this work, we present a family of fuzzy orthogonal projection models for both monolingual and cross-lingual hypernymy prediction. For the monolingual task, we propose a Multi-Wahba Projection (MWP) model to distinguish hypernymy vs. non-hypernymy relations based on word embeddings. This model establishes distributional fuzzy mappings from embeddings of a term to those of its hypernyms and non-hypernyms, which consider the complicated linguistic regularities of these relations. For cross-lingual hypernymy prediction, a Transfer MWP (TMWP) model is proposed to transfer the semantic knowledge from the source language to target languages based on neural word translation. Additionally, an Iterative Transfer MWP (ITMWP) model is built upon TMWP, which augments the training sets of target languages when target languages are lower-resourced with limited training data. Experiments show i) MWP outperforms previous methods over two hypernymy prediction tasks for English; and ii) TMWP and ITMWP are effective to predict hypernymy over seven non-English languages.
{"title":"A Family of Fuzzy Orthogonal Projection Models for Monolingual and Cross-lingual Hypernymy Prediction","authors":"Chengyu Wang, Yan Fan, Xiaofeng He, Aoying Zhou","doi":"10.1145/3308558.3313439","DOIUrl":"https://doi.org/10.1145/3308558.3313439","url":null,"abstract":"Hypernymy is a semantic relation, expressing the “is-a” relation between a concept and its instances. Such relations are building blocks for large-scale taxonomies, ontologies and knowledge graphs. Recently, much progress has been made for hypernymy prediction in English using textual patterns and/or distributional representations. However, applying such techniques to other languages is challenging due to the high language dependency of these methods and the lack of large training datasets of lower-resourced languages. In this work, we present a family of fuzzy orthogonal projection models for both monolingual and cross-lingual hypernymy prediction. For the monolingual task, we propose a Multi-Wahba Projection (MWP) model to distinguish hypernymy vs. non-hypernymy relations based on word embeddings. This model establishes distributional fuzzy mappings from embeddings of a term to those of its hypernyms and non-hypernyms, which consider the complicated linguistic regularities of these relations. For cross-lingual hypernymy prediction, a Transfer MWP (TMWP) model is proposed to transfer the semantic knowledge from the source language to target languages based on neural word translation. Additionally, an Iterative Transfer MWP (ITMWP) model is built upon TMWP, which augments the training sets of target languages when target languages are lower-resourced with limited training data. Experiments show i) MWP outperforms previous methods over two hypernymy prediction tasks for English; and ii) TMWP and ITMWP are effective to predict hypernymy over seven non-English languages.","PeriodicalId":23013,"journal":{"name":"The World Wide Web Conference","volume":"158 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2019-05-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81560714","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Aman Agarwal, Xuanhui Wang, Cheng Li, Michael Bendersky, Marc Najork
Existing unbiased learning-to-rank models use counterfactual inference, notably Inverse Propensity Scoring (IPS), to learn a ranking function from biased click data. They handle the click incompleteness bias, but usually assume that the clicks are noise-free, i.e., a clicked document is always assumed to be relevant. In this paper, we relax this unrealistic assumption and study click noise explicitly in the unbiased learning-to-rank setting. Specifically, we model the noise as the position-dependent trust bias and propose a noise-aware Position-Based Model, named TrustPBM, to better capture user click behavior. We propose an Expectation-Maximization algorithm to estimate both examination and trust bias from click data in TrustPBM. Furthermore, we show that it is difficult to use a pure IPS method to incorporate click noise and thus propose a novel method that combines a Bayes rule application with IPS for unbiased learning-to-rank. We evaluate our proposed methods on three personal search data sets and demonstrate that our proposed model can significantly outperform the existing unbiased learning-to-rank methods.
{"title":"Addressing Trust Bias for Unbiased Learning-to-Rank","authors":"Aman Agarwal, Xuanhui Wang, Cheng Li, Michael Bendersky, Marc Najork","doi":"10.1145/3308558.3313697","DOIUrl":"https://doi.org/10.1145/3308558.3313697","url":null,"abstract":"Existing unbiased learning-to-rank models use counterfactual inference, notably Inverse Propensity Scoring (IPS), to learn a ranking function from biased click data. They handle the click incompleteness bias, but usually assume that the clicks are noise-free, i.e., a clicked document is always assumed to be relevant. In this paper, we relax this unrealistic assumption and study click noise explicitly in the unbiased learning-to-rank setting. Specifically, we model the noise as the position-dependent trust bias and propose a noise-aware Position-Based Model, named TrustPBM, to better capture user click behavior. We propose an Expectation-Maximization algorithm to estimate both examination and trust bias from click data in TrustPBM. Furthermore, we show that it is difficult to use a pure IPS method to incorporate click noise and thus propose a novel method that combines a Bayes rule application with IPS for unbiased learning-to-rank. We evaluate our proposed methods on three personal search data sets and demonstrate that our proposed model can significantly outperform the existing unbiased learning-to-rank methods.","PeriodicalId":23013,"journal":{"name":"The World Wide Web Conference","volume":"11 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2019-05-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81049513","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Ming-Han Feng, Chin-Chi Hsu, Cheng-te Li, Mi-Yen Yeh, Shou-de Lin
Network embedding aims at learning an effective vector transformation for entities in a network. We observe that there are two diverse branches of network embedding: for homogeneous graphs and for multi-relational graphs. This paper then proposes MARINE, a unified embedding framework for both homogeneous and multi-relational networks to preserve both the proximity and relation information. We also extend the framework to incorporate existing features of nodes in a graph, which can further be exploited for the ensemble of embedding. Our solution possesses complexity linear to the number of edges, which is suitable for large-scale network applications. Experiments conducted on several real-world network datasets, along with applications in link prediction and multi-label classification, exhibit the superiority of our proposed MARINE.
{"title":"MARINE: Multi-relational Network Embeddings with Relational Proximity and Node Attributes","authors":"Ming-Han Feng, Chin-Chi Hsu, Cheng-te Li, Mi-Yen Yeh, Shou-de Lin","doi":"10.1145/3308558.3313715","DOIUrl":"https://doi.org/10.1145/3308558.3313715","url":null,"abstract":"Network embedding aims at learning an effective vector transformation for entities in a network. We observe that there are two diverse branches of network embedding: for homogeneous graphs and for multi-relational graphs. This paper then proposes MARINE, a unified embedding framework for both homogeneous and multi-relational networks to preserve both the proximity and relation information. We also extend the framework to incorporate existing features of nodes in a graph, which can further be exploited for the ensemble of embedding. Our solution possesses complexity linear to the number of edges, which is suitable for large-scale network applications. Experiments conducted on several real-world network datasets, along with applications in link prediction and multi-label classification, exhibit the superiority of our proposed MARINE.","PeriodicalId":23013,"journal":{"name":"The World Wide Web Conference","volume":"114 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2019-05-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"83601329","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Alexandra Vtyurina, Adam Fourney, M. Morris, Leah Findlater, Ryen W. White
People with visual impairments often rely on screen readers when interacting with computer systems. Increasingly, these individuals also make extensive use of voice-based virtual assistants (VAs). We conducted a survey of 53 people who are legally blind to identify the strengths and weaknesses of both technologies, as well as the unmet opportunities at their intersection. We learned that virtual assistants are convenient and accessible, but lack the ability to deeply engage with content (e.g., read beyond the first few sentences of Wikipedia), and the ability to get a quick overview of the landscape (list alternative search results & suggestions). In contrast, screen readers allow for deep engagement with content (when content is accessible), and provide fine-grained navigation & control, but at the cost of increased complexity, and reduced walk-up-and-use convenience. In this demonstration, we showcase VERSE, a system that combines the positive aspects of VAs and screen readers, and allows other devices (e.g., smart watches) to serve as optional input accelerators. Together, these features allow people with visual impairments to deeply engage with web content through voice interaction.
{"title":"Bridging Screen Readers and Voice Assistants for Enhanced Eyes-Free Web Search","authors":"Alexandra Vtyurina, Adam Fourney, M. Morris, Leah Findlater, Ryen W. White","doi":"10.1145/3308558.3314136","DOIUrl":"https://doi.org/10.1145/3308558.3314136","url":null,"abstract":"People with visual impairments often rely on screen readers when interacting with computer systems. Increasingly, these individuals also make extensive use of voice-based virtual assistants (VAs). We conducted a survey of 53 people who are legally blind to identify the strengths and weaknesses of both technologies, as well as the unmet opportunities at their intersection. We learned that virtual assistants are convenient and accessible, but lack the ability to deeply engage with content (e.g., read beyond the first few sentences of Wikipedia), and the ability to get a quick overview of the landscape (list alternative search results & suggestions). In contrast, screen readers allow for deep engagement with content (when content is accessible), and provide fine-grained navigation & control, but at the cost of increased complexity, and reduced walk-up-and-use convenience. In this demonstration, we showcase VERSE, a system that combines the positive aspects of VAs and screen readers, and allows other devices (e.g., smart watches) to serve as optional input accelerators. Together, these features allow people with visual impairments to deeply engage with web content through voice interaction.","PeriodicalId":23013,"journal":{"name":"The World Wide Web Conference","volume":"88 4","pages":""},"PeriodicalIF":0.0,"publicationDate":"2019-05-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"91416613","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
When information or infectious diseases spread over a network, in many practical cases, one can observe when nodes adopt information or become infected, but the underlying network is hidden. In this paper, we analyze the problem of finding communities of highly interconnected nodes, given only the infection times of nodes. We propose, analyze, and empirically compare several algorithms for this task. The most stable performance, that improves the current state-of-the-art, is obtained by our proposed heuristic approaches, that are agnostic to a particular graph structure and epidemic model.
{"title":"Learning Clusters through Information Diffusion","authors":"L. Ostroumova, Alexey Tikhonov, N. Litvak","doi":"10.1145/3308558.3313560","DOIUrl":"https://doi.org/10.1145/3308558.3313560","url":null,"abstract":"When information or infectious diseases spread over a network, in many practical cases, one can observe when nodes adopt information or become infected, but the underlying network is hidden. In this paper, we analyze the problem of finding communities of highly interconnected nodes, given only the infection times of nodes. We propose, analyze, and empirically compare several algorithms for this task. The most stable performance, that improves the current state-of-the-art, is obtained by our proposed heuristic approaches, that are agnostic to a particular graph structure and epidemic model.","PeriodicalId":23013,"journal":{"name":"The World Wide Web Conference","volume":"464 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2019-05-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"91478411","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Tiago Santos, Simon Walk, Roman Kern, M. Strohmaier, D. Helic
In this paper, we quantify the impact of self- and cross-excitation on the temporal development of user activity in Stack Exchange Question & Answer (Q&A) communities. We study differences in user excitation between growing and declining Stack Exchange communities, and between those dedicated to STEM and humanities topics by leveraging Hawkes processes. We find that growing communities exhibit early stage, high cross-excitation by a small core of power users reacting to the community as a whole, and strong long-term self-excitation in general and cross-excitation by casual users in particular, suggesting community openness towards less active users. Further, we observe that communities in the humanities exhibit long-term power user cross-excitation, whereas in STEM communities activity is more evenly distributed towards casual user self-excitation. We validate our findings via permutation tests and quantify the impact of these excitation effects with a range of prediction experiments. Our work enables researchers to quantitatively assess the evolution and activity potential of Q&A communities.
{"title":"Self- and Cross-Excitation in Stack Exchange Question & Answer Communities","authors":"Tiago Santos, Simon Walk, Roman Kern, M. Strohmaier, D. Helic","doi":"10.1145/3308558.3313440","DOIUrl":"https://doi.org/10.1145/3308558.3313440","url":null,"abstract":"In this paper, we quantify the impact of self- and cross-excitation on the temporal development of user activity in Stack Exchange Question & Answer (Q&A) communities. We study differences in user excitation between growing and declining Stack Exchange communities, and between those dedicated to STEM and humanities topics by leveraging Hawkes processes. We find that growing communities exhibit early stage, high cross-excitation by a small core of power users reacting to the community as a whole, and strong long-term self-excitation in general and cross-excitation by casual users in particular, suggesting community openness towards less active users. Further, we observe that communities in the humanities exhibit long-term power user cross-excitation, whereas in STEM communities activity is more evenly distributed towards casual user self-excitation. We validate our findings via permutation tests and quantify the impact of these excitation effects with a range of prediction experiments. Our work enables researchers to quantitatively assess the evolution and activity potential of Q&A communities.","PeriodicalId":23013,"journal":{"name":"The World Wide Web Conference","volume":"128 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2019-05-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"84963657","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Personalized Point of Interest (POI) recommendation that incorporates users' personal preferences is an important subject of research. However, challenges exist such as dealing with sparse rating data and spatial location factors. As one of the biggest card payment organizations in the United States, our company holds abundant card payment transaction records with numerous features. In this paper, using restaurant recommendation as a demonstrating example, we present a personalized POI recommendation system (Pcard) that learns user preferences based on user transaction history and restaurants' locations. With a novel embedding approach that captures user embeddings and restaurant embeddings, we model pairwise restaurant preferences with respect to each user based on their locations and dining histories. Finally, a ranking list of restaurants within a spatial region is presented to the user. The evaluation results show that the proposed approach is able to achieve high accuracy and present effective recommendations.
{"title":"Pcard: Personalized Restaurants Recommendation from Card Payment Transaction Records","authors":"Min Du, Robert Christensen, Wei Zhang, Feifei Li","doi":"10.1145/3308558.3313494","DOIUrl":"https://doi.org/10.1145/3308558.3313494","url":null,"abstract":"Personalized Point of Interest (POI) recommendation that incorporates users' personal preferences is an important subject of research. However, challenges exist such as dealing with sparse rating data and spatial location factors. As one of the biggest card payment organizations in the United States, our company holds abundant card payment transaction records with numerous features. In this paper, using restaurant recommendation as a demonstrating example, we present a personalized POI recommendation system (Pcard) that learns user preferences based on user transaction history and restaurants' locations. With a novel embedding approach that captures user embeddings and restaurant embeddings, we model pairwise restaurant preferences with respect to each user based on their locations and dining histories. Finally, a ranking list of restaurants within a spatial region is presented to the user. The evaluation results show that the proposed approach is able to achieve high accuracy and present effective recommendations.","PeriodicalId":23013,"journal":{"name":"The World Wide Web Conference","volume":"65 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2019-05-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"91079781","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Health literacy, i.e. the ability to read and understand medical text, is a relevant component of public health. Unfortunately, many medical texts are hard to grasp by the general population as they are targeted at highly-skilled professionals and use complex language and domain-specific terms. Here, automatic text simplification making text commonly understandable would be very beneficial. However, research and development into medical text simplification is hindered by the lack of openly available training and test corpora which contain complex medical sentences and their aligned simplified versions. In this paper, we introduce such a dataset to aid medical text simplification research. The dataset is created by filtering aligned health sentences using expert knowledge from an existing aligned corpus and a novel simple, language independent monolingual text alignment method. Furthermore, we use the dataset to train a state-of-the-art neural machine translation model, and compare it to a model trained on a general simplification dataset using an automatic evaluation, and an extensive human-expert evaluation.
{"title":"Evaluating Neural Text Simplification in the Medical Domain","authors":"Laurens Van den Bercken, Robert-Jan Sips, C. Lofi","doi":"10.1145/3308558.3313630","DOIUrl":"https://doi.org/10.1145/3308558.3313630","url":null,"abstract":"Health literacy, i.e. the ability to read and understand medical text, is a relevant component of public health. Unfortunately, many medical texts are hard to grasp by the general population as they are targeted at highly-skilled professionals and use complex language and domain-specific terms. Here, automatic text simplification making text commonly understandable would be very beneficial. However, research and development into medical text simplification is hindered by the lack of openly available training and test corpora which contain complex medical sentences and their aligned simplified versions. In this paper, we introduce such a dataset to aid medical text simplification research. The dataset is created by filtering aligned health sentences using expert knowledge from an existing aligned corpus and a novel simple, language independent monolingual text alignment method. Furthermore, we use the dataset to train a state-of-the-art neural machine translation model, and compare it to a model trained on a general simplification dataset using an automatic evaluation, and an extensive human-expert evaluation.","PeriodicalId":23013,"journal":{"name":"The World Wide Web Conference","volume":"1 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2019-05-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"91089628","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Kijung Shin, A. Ghoting, Myunghwan Kim, Hema Raghavan
Given a terabyte-scale graph distributed across multiple machines, how can we summarize it, with much fewer nodes and edges, so that we can restore the original graph exactly or within error bounds? As large-scale graphs are ubiquitous, ranging from web graphs to online social networks, compactly representing graphs becomes important to efficiently store and process them. Given a graph, graph summarization aims to find its compact representation consisting of (a) a summary graph where the nodes are disjoint sets of nodes in the input graph, and each edge indicates the edges between all pairs of nodes in the two sets; and (b) edge corrections for restoring the input graph from the summary graph exactly or within error bounds. Although graph summarization is a widely-used graph-compression technique readily combinable with other techniques, existing algorithms for graph summarization are not satisfactory in terms of speed or compactness of outputs. More importantly, they assume that the input graph is small enough to fit in main memory. In this work, we propose SWeG, a fast parallel algorithm for summarizing graphs with compact representations. SWeG is designed for not only shared-memory but also MapReduce settings to summarize graphs that are too large to fit in main memory. We demonstrate that SWeG is (a) Fast: SWeG is up to 5400 × faster than its competitors that give similarly compact representations, (b) Scalable: SWeG scales to graphs with tens of billions of edges, and (c) Compact: combined with state-of-the-art compression methods, SWeG achieves up to 3.4 × better compression than them.
{"title":"SWeG: Lossless and Lossy Summarization of Web-Scale Graphs","authors":"Kijung Shin, A. Ghoting, Myunghwan Kim, Hema Raghavan","doi":"10.1145/3308558.3313402","DOIUrl":"https://doi.org/10.1145/3308558.3313402","url":null,"abstract":"Given a terabyte-scale graph distributed across multiple machines, how can we summarize it, with much fewer nodes and edges, so that we can restore the original graph exactly or within error bounds? As large-scale graphs are ubiquitous, ranging from web graphs to online social networks, compactly representing graphs becomes important to efficiently store and process them. Given a graph, graph summarization aims to find its compact representation consisting of (a) a summary graph where the nodes are disjoint sets of nodes in the input graph, and each edge indicates the edges between all pairs of nodes in the two sets; and (b) edge corrections for restoring the input graph from the summary graph exactly or within error bounds. Although graph summarization is a widely-used graph-compression technique readily combinable with other techniques, existing algorithms for graph summarization are not satisfactory in terms of speed or compactness of outputs. More importantly, they assume that the input graph is small enough to fit in main memory. In this work, we propose SWeG, a fast parallel algorithm for summarizing graphs with compact representations. SWeG is designed for not only shared-memory but also MapReduce settings to summarize graphs that are too large to fit in main memory. We demonstrate that SWeG is (a) Fast: SWeG is up to 5400 × faster than its competitors that give similarly compact representations, (b) Scalable: SWeG scales to graphs with tens of billions of edges, and (c) Compact: combined with state-of-the-art compression methods, SWeG achieves up to 3.4 × better compression than them.","PeriodicalId":23013,"journal":{"name":"The World Wide Web Conference","volume":"53 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2019-05-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"78126246","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Given posts on 'abortion' and posts on 'religion' from a political forum, how can we find topics that are discriminative and those in common? In general, (1) how can we compare and contrast two or more different ('labeled') document collections? Moreover, (2) how can we visualize the data (in 2-d or 3-d) to best reflect the similarities and differences between the collections? We introduce (to the best of our knowledge) the first contrastive and visual topic model, called ContraVis, that jointly addresses both problems: (1) contrastive topic modeling, and (2) contrastive visualization. That is, ContraVis learns not only latent topics but also embeddings for the documents, topics and labels for visualization. ContraVis exhibits three key properties by design. It is (i) Contrastive: It enables comparative analysis of different document corpora by extracting latent discriminative and common topics across labeled documents; (ii) Visually-expressive: Different from numerous existing models, it also produces a visualization for all of the documents, labels, and the extracted topics, where proximity in the coordinate space is reflective of proximity in semantic space; (iii) Unified: It extracts topics and visual coordinates simultaneously under a joint model. Through extensive experiments on real-world datasets, we show ContraVis 's potential for providing visual contrastive analysis of multiple document collections. We show both qualitatively and quantitatively that ContraVis significantly outperforms both unsupervised and supervised state-of-the-art topic models in contrastive power, semantic coherence and visual effectiveness.
{"title":"ContraVis: Contrastive and Visual Topic Modeling for Comparing Document Collections","authors":"T. Le, L. Akoglu","doi":"10.1145/3308558.3313617","DOIUrl":"https://doi.org/10.1145/3308558.3313617","url":null,"abstract":"Given posts on 'abortion' and posts on 'religion' from a political forum, how can we find topics that are discriminative and those in common? In general, (1) how can we compare and contrast two or more different ('labeled') document collections? Moreover, (2) how can we visualize the data (in 2-d or 3-d) to best reflect the similarities and differences between the collections? We introduce (to the best of our knowledge) the first contrastive and visual topic model, called ContraVis, that jointly addresses both problems: (1) contrastive topic modeling, and (2) contrastive visualization. That is, ContraVis learns not only latent topics but also embeddings for the documents, topics and labels for visualization. ContraVis exhibits three key properties by design. It is (i) Contrastive: It enables comparative analysis of different document corpora by extracting latent discriminative and common topics across labeled documents; (ii) Visually-expressive: Different from numerous existing models, it also produces a visualization for all of the documents, labels, and the extracted topics, where proximity in the coordinate space is reflective of proximity in semantic space; (iii) Unified: It extracts topics and visual coordinates simultaneously under a joint model. Through extensive experiments on real-world datasets, we show ContraVis 's potential for providing visual contrastive analysis of multiple document collections. We show both qualitatively and quantitatively that ContraVis significantly outperforms both unsupervised and supervised state-of-the-art topic models in contrastive power, semantic coherence and visual effectiveness.","PeriodicalId":23013,"journal":{"name":"The World Wide Web Conference","volume":"27 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2019-05-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81912831","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：